top of page
Search
Writer's pictureDavid Kurtenbach
Oct 16, 20213 min read

Deep Learning Applications: Medical Diagnosis thorough Image Detection

Artificial intelligence and deep learning algorithms have a wide variety of applications in today's world. While most people might not be able to readily point out those use cases, they are encountered on a daily basis in dozens of different scenarios. Your recommended shows list on Netflix, Google searches that match people with the exactly what they are looking for, or voice recognition when talking to Alexa are all examples of AI.


Example of image detection where a family on the beach is recognized and interpreted by the AI as people flying a kite

One particularly powerful application of AI and machine learning (ML) is image recognition. Image recognition in AI is a tool where a computer can "look" at a picture or a image, interpret what's in the image, and categorize or classify it accordingly. In some regards, classification, sometimes called grouping or clustering, lies at the very core of AI and ML. Image recognition is similar to how people understand what's in a picture. Looking at the image on the right, think about how you would describe and how your brain is able to know there are people running on the beach versus dogs playing in an open field. Image recognition breaks down the images into a language that the computer program can understand then identifies the patterns to ultimately classify what is happening in the picture.


At face value, image recognition sounds like an amazing advancement in technology but it has been around since the 60's and continues to improve. The real value of this type of technology comes in how it is used. One field that has found a way to use deep learning image recognition is the world of medical diagnosis. Medical imaging is an area that stands to gain an untold amount of value from deep learning image recognition. X-rays are one of the best ways for a medical professional to tell what is happening with a patient, it's the old adage of a picture is worth a thousand words. Patients can describe their symptoms to a doctor but an x-ray can effectively confirm the problem. A major bottleneck to quickly examining a stack of x-rays is the need for human interpretation of each one. Image recognition has been employed to help resolve those challenges.


Multiple MRI brain scans

One specific example is the Deep Patient experiment. Researchers were looking for a way to quickly diagnosis MRI scans of patients as having schizophrenia or not. Medical professionals were able to identify criteria commonly scene in people with schizophrenia and able to develop an image recognition tool that could identify and diagnose. This is key to using AI and ML, programmers and professionals need to clearly identify qualifying criteria for the AI to recognize. In the case of Deep Patient, common characteristics of a person with schizophrenia include overall gray matter loss, decreased volume of bilateral medial temporal areas, and a left superior temporal region deficit. (Jihoon, O., Baek-Lok, O., Kyong-Uk, L., Jeong-Ho, C., & Kyongsik, Y, 2020) Researchers established a training data set using existing data and were able to develop a tool that was 97% accurate in diagnosis MRI scans of people with schizophrenia. It's not just MRIs, but the same has been done to varying degrees of accuracy for glaucoma, lung cancer, and diabetic retinopathy. (Aggarwal, R., Sounderajah, V., Martin, G., Ting, D., Karthikesalingam, A., King, D., Ashrafain, Hu. & Darzi, A., 2021)


As technology advances more and more the use of AI, ML, deep learning, and big data will grow exponentially. The understanding and applications of it's value are just beginning to used in real-world scenarios. Examples like Deep Patient and medical imaging prove that people stand to gain a lot from embracing new technologies.


REFERENCES;

49 views0 comments

Comments

bottom of page